Run DeepSeek Locally on Mac: A Step-by-Step Guide

Running DeepSeek locally lets you chat with a downloaded language model on your Mac instead of sending each prompt to a hosted chat service. You’ll install a local model runner, choose a model that fits your Mac’s memory, and open a chat in your browser; an internet connection is needed for the initial download.

What You’ll Need

A Mac with macOS and enough free storage for the model. A smaller, quantized model is more practical on Macs with limited unified memory.
An internet connection to install the runner and download model files.
An administrator account to install software, plus basic familiarity with Terminal.
Optional: Apple silicon generally runs supported models efficiently, but the exact speed depends on the Mac, model size, and settings. Intel Macs may be slower.

Step-by-Step

Check your Mac’s memory and free space. Open the Apple menu, choose About This Mac, and note the memory or unified memory. Check available storage in System Settings > General > Storage. Model downloads can take several gigabytes or more; leave room for the app, model, and other files. A model’s listed file size is not a guarantee of its full working-memory requirement.
Choose a local runner. This guide uses Ollama, a local model runner with a macOS app and Terminal commands. Go to its official website, download the macOS version, open the downloaded file, and follow the installer instructions. If macOS asks whether to open the app, confirm only if you downloaded it from the official source and recognize the app. The app may ask for permission to install its command-line component; approve that request to use the commands below.
Confirm the runner is available. Open Terminal from Applications > Utilities. Enter ollama --version and press Return. If it prints a version, continue. If Terminal says the command cannot be found, launch the Ollama app from Applications, allow its setup to finish, then open a new Terminal window and try again.
Select a DeepSeek model. Check the runner’s model library for available DeepSeek models and their exact names and tags. DeepSeek offers different model families and sizes; some are general-purpose, while others are designed for particular tasks. Pick a smaller instruct/chat model if you mainly want conversational use or have limited memory. Read the model’s description and requirements rather than assuming that every DeepSeek model has the same capabilities.
Download and start the model. In Terminal, run ollama run MODEL_NAME, replacing MODEL_NAME with the exact name and tag shown in the model library. For example, use the library’s current DeepSeek model name rather than copying a potentially outdated tag from an old tutorial. Ollama downloads the model on the first run, then starts an interactive prompt. The download can take time; keep the Mac awake and connected to the internet.
Send a test prompt. When the prompt appears, type a simple request such as Explain how photosynthesis works in three sentences. Press Return and wait for the response. You can ask follow-up questions in the same session. To leave the chat, enter /bye or use Control-C in Terminal.
Use the browser interface, if preferred. The Ollama app can provide a local service for compatible chat interfaces, but the runner alone may not include a full chat website. Install a reputable interface that explicitly supports Ollama, follow its official setup steps, and configure its connection to the local Ollama service. Do not assume a web page is local just because it looks like a chat app; check that it is connecting to your Mac’s local service and avoid entering sensitive information into untrusted software.
Manage models and updates. To see downloaded models, run ollama list. To remove one you no longer need, run ollama rm MODEL_NAME with its exact listed name. Update the runner by downloading and installing the current release from its official source. Check the model library for a newer model tag; updating the app does not necessarily replace models already downloaded.

Common Mistakes

Choosing a model that is too large. A model may download successfully but run slowly, use swap, or fail to load. Start with a smaller model and move up only if memory and speed are acceptable.
Using the wrong model name. Names and tags can change. Copy the exact current identifier from the runner’s model library; a mistyped or obsolete tag can produce a “not found” error.
Expecting hosted-service performance. Local generation speed depends on the Mac and model. Large models can respond slowly, especially on machines with less memory or older processors.
Assuming every answer is reliable. Local operation does not make a model accurate. Verify important facts, calculations, and code, and do not treat its output as professional advice.
Confusing a local runner with a chat interface. The runner downloads and serves models; you may need a separate compatible app for a graphical chat window.

Tips

Begin with one model and a short prompt. This makes it easier to tell whether a problem is caused by setup, memory, or the model itself.
Use a model intended for instruction following for ordinary question-and-answer chat. Check its description for language and task strengths.
Keep the Mac connected to power for long downloads or sessions, and close memory-heavy apps if generation is sluggish.
Local use can reduce the need to send prompts to a hosted service, but it does not automatically guarantee privacy. Check the runner and any interface for network features, logging, and data handling.
For repeat use, save useful prompts in a text file. Avoid pasting passwords, private keys, or confidential material into any model unless you understand the software’s data handling.

Troubleshooting

“Command not found” in Terminal: Open the installed Ollama app, wait for setup, then open a fresh Terminal window. If it persists, reinstall from the official download and follow any prompt to install the command-line component.

Model name not found: Check the library for the exact name and tag, including punctuation, then retry. Confirm your internet connection if the runner cannot retrieve the model listing or files.

Download stalls or fails: Check the connection and available disk space, then retry the same run command. Large downloads can take a while. If you use a VPN, proxy, or managed network, it may block downloads; check its settings or try a permitted connection.

Model is slow, Mac heats up, or memory warnings appear: Stop the session, quit unused apps, and try a smaller model. Avoid repeatedly starting a model that exceeds available memory. A restart can clear memory pressure, but it cannot make an oversized model fit.

No browser chat appears: The command-line chat is the basic interface. For a graphical interface, install a compatible client and verify its setup instructions and local connection. If Terminal cannot connect to the runner, open the Ollama app and check that it is running, then retry.

Frequently Asked Questions

Can I run DeepSeek on a Mac without internet?
After the runner and model have been downloaded, basic local inference can work offline. Internet is needed for downloads, updates, and any optional online features.

Does this work on both Apple silicon and Intel Macs?
Ollama supports macOS, but performance and model compatibility vary. Apple silicon is generally a better fit for local inference; check current requirements for the runner and chosen model.

How much memory do I need?
There is no single amount for every DeepSeek model. Larger models and longer conversations need more memory. Check the model’s current size and requirements, and start small if unsure.

Is DeepSeek free to run locally?
The runner and model may be available to download without a fee, but storage, electricity, and your Mac are still required. Check the current license and terms for the specific model before using it.

Are my prompts automatically private?
Local inference keeps model generation on your Mac, but privacy also depends on the runner, chat interface, and any enabled network features. Review their documentation and avoid untrusted clients.

Can I use the same model in other apps?
Some apps can connect to Ollama’s local service. Compatibility depends on the app and model; follow the app’s instructions and use the exact connection settings it specifies.

General steps for current mainstream versions; menus can differ by device and version. See the disclaimer or report an error.

More in this bed

The whole Tech How-To bed →